[racket] disable garbage collection

From: Neil Van Dyke (neil at neilvandyke.org)
Date: Thu Jun 2 12:24:12 EDT 2011

My current interest is in satisfying a client's request (to analyze 
certain things with the GC disabled), but here's a few reasons I can 
imagine that people sometimes want to disable GC in languages in general 
(not specific to Racket):

* Performance: you have a fairly short-lived performance-sensitive 
program, you don't want to pay the time cost of GC, and you have enough 
RAM.  In Racket, I haven't noticed GC being a problem.  (Before Java 
became popular, and when computers were a *lot* slower, it was common 
for people to be prejudiced against GC'd languages, and there were jokes 
about getting "collecting garbage..." messages at inopportune moments in 
life-critical real-time systems.  I don't see much of that sentiment 
anymore.)

* Evaluation: related to performance, but another way to measure the 
cost of GC when tuning, with the intention of normally running with GC.  
In Racket, there are other ways to measure GC cost, and this would be an 
additional way.

* Debugging: if you're debugging when doing low-level programming or 
interfacing with less-trustworthy C code, disabling GC lets you 
eliminate GC as a possible cause of trouble.  Again, in Racket, this 
isn't a concern for most people, just like it's not a concern for most 
Java programmers.

* Extremely Unusual Situations: if you're in the unusual situation of 
doing something normally considered inadvisable, like you're 
intentionally designing the SR-71 to leak fuel, maybe there's some 
bizarre rocket-scientist reason understood only by you that you have to 
turn off GC just to get your app to fly in one piece.  Almost no one 
will ever need this, and, in the case of Racket, they could always add 
it if they did.

BTW, for the mechanism for disabling GC, I think I actually have a 
preference for having all the GC code in place as normal, so that the 
usual code paths are used, and simply having the (conceptual) 
do-we-want-to-do-a-gc-cycle-now function always return false.  I'd try 
to avoid running any blocks of code not already covered by users who are 
running GC.


Eduardo Bellani wrote at 06/02/2011 10:01 AM:
> Out of curiosity, why would you want to disable GC?
>
> On 06/02/2011 10:30 AM, Matthew Flatt wrote:
>   
>> At Thu, 02 Jun 2011 00:25:04 -0400, Neil Van Dyke wrote:
>>     
>>> When running Racket apps from the command line, is there a way to 
>>> disable garbage collection?
>>>       
>> There's currently no way to disable GC. I'll look into adding one.
>>     

-- 
http://www.neilvandyke.org/


Posted on the users mailing list.